Earthquake surface ruptures mornitoring based on deep learning and remote sensing

Surface expression of seismic rupture is a distinctive feature of large earthquakes, and the length of the surface rupture varies from several to several hundred kilometers depending on the magnitude of the earthquake. The structure and mechanical properties of fault zones strongly influence the behavior of earthquake ruptures, and detailed mapping and documentation of fault geometry such as fault bends, steps, branches, and their related segment geometry play a crucial role in determining the propagation and path of earthquake ruptures. The most precise method to map earthquake surface ruptures in detail might be manual mapping in the field by an expert; however, it takes a long time to analyze a vast area, and not everyone has access to the necessary expertise. As an alternative, rupture mapping based on remote sensing has been proposed to supplement the limitations of these field surveys.

In this study, we propose a robust model for automatic detection and analysis of morphological features of surface ruptures using deep learning based on previous research results. Additionally, a skeletonization technique was developed for morphological analysis of detected fractures, and fractures of complex structures on the ground were objectified as individual fractures. Finally, the geometric quantitative characteristics of individualized fractures, including crack location, length, width, and direction, were automatically extracted. By comparing the detection results of the proposed model with the ground truth confirmed by the expert using a line map, the reliability of the entire model could be confirmed.

To examine the applicability of the proposed model, the detection performance of various surface ruptures in localized areas was evaluated, and key characteristics, including rupture direction and pattern for extensive surface ruptures, were clearly identified. The satellite image-based surface destruction detection model proposed in this study can be used as a useful tool for field investigation and earthquake-related basic data collection by automatically detecting various surface destruction and deformations caused by earthquakes. Therefore, the proposed model, which enables fast and accurate fracture and fault mapping and quantitative analysis using high-resolution satellite data, is expected to be utilized as an important integrated solution. It would also be extremely beneficial in characterizing and comprehending the structural, geometrical complexity, and mechanical properties of fault zones.